A case-informed paper · Uppsala University · 2026
Dealing with AI in Higher Education – How to go from theory to practice?
Generative AI is already part of academic work. The practical question is how universities can guide its use while preserving intellectual effort, transparent authorship and credible assessment.
Anders Randler & Raine Isaksson · Uppsala University, Campus Gotland · Case-informed synthesis
The institutional question
Prohibition leaves actual practice hidden. Passive acceptance leaves it unguided. The paper organises a response around three questions.
How can teachers and students respond when AI tools change faster than ordinary planning cycles?
How can AI be incorporated transparently without surrendering academic judgment?
How should education protect cognitive development and verify independent understanding?
What the evidence suggests
A 19-study meta-analysis cited by the paper reports markedly different outcomes depending on whether educators scaffold GenAI use.
The educational variable is not simply whether students use AI. It is whether educators design the interaction, frame its purpose and require students to evaluate what the system produces. Randler & Isaksson, 2026; Liu et al., 2025
Reported overall effect: g = 0.683. Effect sizes describe the cited evidence base, not an expected result for every course.
The three-pillar model
The model works as a dependency chain. Faculty competence enables transparent integration; assessment protects the learning that integration is meant to support.
Foundation
Educators need technical, pedagogical and ethical fluency to recognise unreliable output, design meaningful AI-supported tasks and explain their decisions to students.
In practice: provide hands-on development connected to real teaching, research and administrative work.
Application
Replace “shadow use” with explicit syllabus guidance and credit-bearing AI literacy education. Students should learn to question, edit, source-check and disclose—not merely prompt.
In practice: state where AI is required, permitted, restricted or excluded, and connect each decision to learning outcomes.
Safeguard
AI can improve a visible product while bypassing the cognitive work that makes knowledge durable. Learning may be open and AI-supported; assessment must still verify individual understanding.
In practice: combine documented AI-supported work with oral, practical or process-based evidence of reasoning.
Theory into practice · Campus Gotland
The departmental case combines a student course with targeted faculty development. It is an implementation example, not a controlled trial.
For students · 5 credits
For faculty & researchers · 3 modules
Three hands-on modules of three to five hours each establish a shared vocabulary for institutional practice.
Preserving desirable difficulties
“AI should be open in learning and accountable in assessment.”
Learning phase
Students may use AI for explanations, critique, literature exploration, debugging and alternative perspectives. Use is documented, discussed and source-checked.
Assessment phase
Students demonstrate what they understand through structured oral discussion, problem solving, practical work or other direct evidence.
Important: oral assessment is not presented as a universal replacement for written work. Validity and equity require structured questions, explicit rubrics, examiner calibration, accommodations and complementary evidence.
Evidence boundary
The paper combines research synthesis and policy guidance with early experience from one department. It does not report a controlled evaluation of the local initiatives.
The paper also discloses how AI supported research, reading, drafting and refinement—putting its own transparency principle into practice.
About the author
University lecturer at Uppsala University, Campus Gotland, teaching mathematics and artificial intelligence.
For more than a decade, he has also educated teachers through his role as an Apple Distinguished Educator. His work connects technological change with the practical design of teaching, learning and assessment.
Read Anders’ profileContinue reading
The full paper contains the methodology, departmental case, assessment proposals, limitations and references in their original context.